Quarantine stale service endpoints on bootstrap

[kacy]

Summary:

• quarantine durable service endpoints whose node ids no longer exist in the cluster agent table
• rerun authoritative bootstrap after the cluster DB is attached so cluster-mode startup can drain stale rows before rebuilding DNS and compatibility state
• expose stale-endpoint quarantine counters in rollout status and Prometheus metrics, with a test covering the boot path

Verification:

• zig fmt on touched files
• timeout 30 env YOQ_SKIP_SLOW_TESTS=1 ZIG_GLOBAL_CACHE_DIR=.zig-global-cache ZIG_LOCAL_CACHE_DIR=.zig-local-cache zig build test -- --test-filter bootstrap quarantines stale endpoint rows for missing nodes (timed out with no emitted failures)
• timeout 30 env YOQ_SKIP_SLOW_TESTS=1 ZIG_GLOBAL_CACHE_DIR=.zig-global-cache ZIG_LOCAL_CACHE_DIR=.zig-local-cache zig build test -- --test-filter route handles /v1/status?mode=service_rollout GET (timed out with no emitted failures)
• timeout 30 env YOQ_SKIP_SLOW_TESTS=1 ZIG_GLOBAL_CACHE_DIR=.zig-global-cache ZIG_LOCAL_CACHE_DIR=.zig-local-cache zig build test -- --test-filter handleMetricsPrometheus exposes service rollout metrics (timed out with no emitted failures)